Cast shorthand in record expression creates field name "cast"

[philrz]

tl;dr

As a user, the field name cast seems unexpected here.

$ echo '{x:1, y:2}' | zq 'cut x,float64(y)' -
{x:1,cast:2.}

Details

Repro is with Zed commit b8fe312.

In the language, field names are commonly derived from function names when the calls are made in record expressions like this. For instance:

$ zq -version
Version: v1.18.0-11-gb8fe3121

$ echo '{x:1, y:-2}' | zq 'cut x,abs(y)' -
{x:1,abs:2}

And of course, if a user wants to explicitly assign a field name to avoid this effect, they can include explicit assignment.

$ echo '{x:1, y:-2}' | zq 'cut x,y:=abs(y)' -
{x:1,y:2}

That said, going back to our first example, the field name cast would only make sense if the user knows the implementation detail of float64() being a wrapper around the cast() function. Instead one of the following would seem more intuitive:

1. Using the field name float64 to match the behavior we just saw with abs().
2. Maintaining the original field name (y in this case).

In a group discussion, there was consensus on maintaining the original field name, which happens to match my instincts as well.

In that same discussion, @mattnibs pointed out a related surprise that when dealing with nested fields, this casting operation creates a top-level field, whereas he's felt it should keep the value in its nested path location.

$ echo '{x:1, y:2, a: {b: 3}}' | zq 'cut x,float64(a.b)' -
{x:1,cast:3.}

If we don't address that when taking up this issue, I can open a separate issue at that time.